Re: [qemu-s390x] [PATCH RFC] s390/css: handle CCW_FLAG_SKIP

[Cornelia Huck]

On Mon, 6 May 2019 16:58:03 -0400
Eric Farman <address@hidden> wrote:

> On 5/6/19 1:17 PM, Cornelia Huck wrote:
> > If a ccw has CCW_FLAG_SKIP set, and the command is of type
> > read, read backwards, or sense, no data should be written
> > to the guest for that command.
> > 
> > Signed-off-by: Cornelia Huck <address@hidden>
> > ---
> > 
> > Only extremely lightly tested (i.e., can boot a guest.)
> > 
> > ---
> >   hw/s390x/css.c         | 22 ++++++++++++++++++----
> >   include/hw/s390x/css.h |  1 +
> >   2 files changed, 19 insertions(+), 4 deletions(-)
> > 
> > diff --git a/hw/s390x/css.c b/hw/s390x/css.c
> > index 8fc9e35ba5d3..6ee31cc2e08f 100644
> > --- a/hw/s390x/css.c
> > +++ b/hw/s390x/css.c
> > @@ -830,8 +830,12 @@ static int ccw_dstream_rw_noflags(CcwDataStream *cds, 
> > void *buff, int len,
> >       if (op == CDS_OP_A) {
> >           goto incr;
> >       }
> > -    ret = address_space_rw(&address_space_memory, cds->cda,
> > -                           MEMTXATTRS_UNSPECIFIED, buff, len, op);
> > +    if (!cds->do_skip) {
> > +        ret = address_space_rw(&address_space_memory, cds->cda,
> > +                               MEMTXATTRS_UNSPECIFIED, buff, len, op);
> > +    } else {
> > +        ret = 0;
> > +    }
> >       if (ret != MEMTX_OK) {
> >           cds->flags |= CDS_F_STREAM_BROKEN;
> >           return -EINVAL;
> > @@ -928,8 +932,13 @@ static int ccw_dstream_rw_ida(CcwDataStream *cds, void 
> > *buff, int len,
> >       do {
> >           iter_len = MIN(len, cont_left);
> >           if (op != CDS_OP_A) {
> > -            ret = address_space_rw(&address_space_memory, cds->cda,
> > -                                   MEMTXATTRS_UNSPECIFIED, buff, iter_len, 
> > op);
> > +            if (!cds->do_skip) {
> > +                ret = address_space_rw(&address_space_memory, cds->cda,
> > +                                       MEMTXATTRS_UNSPECIFIED, buff, 
> > iter_len,
> > +                                       op);
> > +            } else {
> > +                ret = 0;
> > +            }
> >               if (ret != MEMTX_OK) {
> >                   /* assume inaccessible address */
> >                   ret = -EINVAL; /* channel program check */
> > @@ -968,6 +977,11 @@ void ccw_dstream_init(CcwDataStream *cds, CCW1 const 
> > *ccw, ORB const *orb)
> >   
> >       cds->count = ccw->count;
> >       cds->cda_orig = ccw->cda;
> > +    /* skip is only effective for read, read backwards, or sense commands 
> > */
> > +    cds->do_skip = (ccw->flags & CCW_FLAG_SKIP) &&
> > +        (ccw->cmd_code & CCW_CMD_BASIC_SENSE ||
> > +         ccw->cmd_code & 0x02 /* read */ ||
> > +         ccw->cmd_code & 0x0c /* read backwards */);  
> 
> I wish so badly that these checks work, since it'd simplify the vfio-ccw 
> code, but I don't think this lets you tell the difference between a READ 
> (x02) and a NOP (x03) or any other control-type CCW.  Ditto 
> read-backward versus TIC.  :-(

Drat, why is that never easy... I'll rework this check.

> 
> >       ccw_dstream_rewind(cds);
> >       if (!(cds->flags & CDS_F_IDA)) {
> >           cds->op_handler = ccw_dstream_rw_noflags;
> > diff --git a/include/hw/s390x/css.h b/include/hw/s390x/css.h
> > index aae19c427229..7cc183ef4366 100644
> > --- a/include/hw/s390x/css.h
> > +++ b/include/hw/s390x/css.h
> > @@ -97,6 +97,7 @@ typedef struct CcwDataStream {
> >       int (*op_handler)(struct CcwDataStream *cds, void *buff, int len,
> >                         CcwDataStreamOp op);
> >       hwaddr cda;
> > +    bool do_skip;
> >   } CcwDataStream;
> >   
> >   /*
> >   
>